--- license: apache-2.0 datasets: - ai2_arc - allenai/ultrafeedback_binarized_cleaned - argilla/distilabel-intel-orca-dpo-pairs - jondurbin/airoboros-3.2 - codeparrot/apps - facebook/belebele - bluemoon-fandom-1-1-rp-cleaned - boolq - camel-ai/biology - camel-ai/chemistry - camel-ai/math - camel-ai/physics - jondurbin/contextual-dpo-v0.1 - jondurbin/gutenberg-dpo-v0.1 - jondurbin/py-dpo-v0.1 - jondurbin/truthy-dpo-v0.1 - LDJnr/Capybara - jondurbin/cinematika-v0.1 - WizardLM/WizardLM_evol_instruct_70k - glaiveai/glaive-function-calling-v2 - jondurbin/gutenberg-dpo-v0.1 - grimulkan/LimaRP-augmented - lmsys/lmsys-chat-1m - ParisNeo/lollms_aware_dataset - TIGER-Lab/MathInstruct - Muennighoff/natural-instructions - openbookqa - kingbri/PIPPA-shareGPT - piqa - Vezora/Tested-22k-Python-Alpaca - ropes - cakiki/rosetta-code - Open-Orca/SlimOrca - b-mc2/sql-create-context - squad_v2 - mattpscott/airoboros-summarization - migtissera/Synthia-v1.3 - unalignment/toxic-dpo-v0.2 - WhiteRabbitNeo/WRN-Chapter-1 - WhiteRabbitNeo/WRN-Chapter-2 - winogrande --- # A bagel, with everything (except DPO) ![bagel](bagel.png) ## Overview This is the pre-DPO version of the mistral-7b model fine-tuned with https://github.com/jondurbin/bagel The DPO counterpart can be found here: https://huggingface.co/jondurbin/bagel-dpo-7b-v0.4 This model is likely better for roleplay usage. ### Data sources There are many data sources used in the bagel models. See https://github.com/jondurbin/bagel for more information. __*Only train splits are used, and a decontamination by cosine similarity is performed at the end as a sanity check against common benchmarks. If you don't know the difference between train and test, please learn.*__
SFT data sources - [ai2_arc](https://huggingface.co/datasets/ai2_arc) - Abstraction and reasoning dataset, useful in measuring "intelligence" to a certain extent. - [airoboros](https://huggingface.co/datasets/unalignment/spicy-3.1) - Variety of categories of synthetic instructions generated by gpt-4. - [apps](https://huggingface.co/datasets/codeparrot/apps) - Python coding dataset with 10k problems. - [belebele](https://huggingface.co/datasets/facebook/belebele) - Multi-lingual reading comprehension dataset. - [bluemoon](https://huggingface.co/datasets/Squish42/bluemoon-fandom-1-1-rp-cleaned) - Roleplay data scraped from Bluemoon, then cleaned and formatted as ShareGPT. - [boolq](https://huggingface.co/datasets/boolq) - Corpus of yes/no questions (which can be surprisingly difficult for AI to answer apparently?) - [camel-ai biology](https://huggingface.co/datasets/camel-ai/biology) - GPT-4 generated biology instructions. - [camel-ai chemistry](https://huggingface.co/datasets/camel-ai/chemistry) - GPT-4 generated chemistryinstructions. - [camel-ai math](https://huggingface.co/datasets/camel-ai/math) - GPT-4 generated math instructions. - [camel-ai physics](https://huggingface.co/datasets/camel-ai/physics) - GPT-4 generated physics instructions. - [capybara](https://huggingface.co/datasets/LDJnr/Capybara) - Multi-turn dataset used to create the capybara models. - [cinematika](https://huggingface.co/datasets/jondurbin/cinematika-v0.1) (instruction and plain text) - RP-style data synthesized from movie scripts so the model isn't quite as boring as it otherwise would be. - [emobank](https://github.com/JULIELab/EmoBank) - Emotion annotations using the Valence-Arousal-Domninance scheme. - [evol-instruct](https://huggingface.co/datasets/WizardLM/WizardLM_evol_instruct_70k) - WizardLM's evol instruct 70k dataset. - [glaive-function-calling-v2](https://huggingface.co/datasets/glaiveai/glaive-function-calling-v2) - GlaiveAI function calling dataset. - [gutenberg](https://www.gutenberg.org/) (plain text) - Books/plain text, again to make the model less boring, only a handful of examples supported by [chapterize](https://github.com/JonathanReeve/chapterize) - [limarp-augmented](https://huggingface.co/datasets/grimulkan/LimaRP-augmented) - Augmented and further modified version of [LimaRP](https://huggingface.co/datasets/lemonilia/LimaRP) - [lmsys_chat_1m](https://huggingface.co/datasets/lmsys/lmsys-chat-1m) (only gpt-4 items, also used for DPO) - Chats collected by the lmsys chat arena, containing a wide variety of chats with various models. - [lollms](https://huggingface.co/datasets/ParisNeo/lollms_aware_dataset) - LoLLMs question answering dataset by ParisNeo, with helpful question answer pairs for using LoLLMs. - [mathinstruct](https://huggingface.co/datasets/TIGER-Lab/MathInstruct) - Composite dataset with a variety of math-related tasks and problem/question formats. - [natural_instructions](https://huggingface.co/datasets/Muennighoff/natural-instructions) - Millions of instructions from 1600+ task categories (sampled down substantially, stratified by task type) - [openbookqa](https://huggingface.co/datasets/openbookqa) - Question answering dataset. - [pippa](https://huggingface.co/datasets/kingbri/PIPPA-shareGPT) - Deduped version of [PIPPA](https://huggingface.co/datasets/PygmalionAI/PIPPA) in ShareGPT format. - [piqa](https://huggingface.co/datasets/piqa) - Phyiscal interaction question answering. - [python_alpaca](https://huggingface.co/datasets/Vezora/Tested-22k-Python-Alpaca) - Python instruction response pairs, validated as functional. - [ropes](https://huggingface.co/datasets/ropes) - Reasoning Over PAragraph Effects in Situations - enhances ability to apply knowledge from a passage of text to a new situation. - [rosetta_code](https://huggingface.co/datasets/cakiki/rosetta-code) - Code problems and solutions in a variety of programming languages taken from rosettacode.org. - [slimorca](https://huggingface.co/datasets/Open-Orca/SlimOrca) - Collection of ~500k gpt-4 verified chats from OpenOrca. - [sql-create-context](https://huggingface.co/datasets/b-mc2/sql-create-context) - SQL-targeted dataset, combining WikiSQL and Spider. - [squad_v2](https://huggingface.co/datasets/squad_v2) - Contextual question answering (RAG). - [airoboros-summarization](https://huggingface.co/datasets/mattpscott/airoboros-summarization) - Combination of various summarization datasets, formatted into the airoboros context-obedient format. - [synthia](https://huggingface.co/datasets/migtissera/Synthia-v1.3) - GPT-4 generated data using advanced prompting from Migel Tissera. - whiterabbitneo [chapter 1](https://huggingface.co/datasets/WhiteRabbitNeo/WRN-Chapter-1) and [chapter 2](https://huggingface.co/datasets/WhiteRabbitNeo/WRN-Chapter-2) - Offensive cybersecurity dataset by WhiteRabbitNeo/Migel Tissera - [winogrande](https://huggingface.co/datasets/winogrande) - Fill in the blank style prompts.
DPO data sources - [airoboros 3.2](https://huggingface.co/datasets/jondurbin/airoboros-3.2) vs [airoboros m2.0](https://huggingface.co/datasets/jondurbin/airoboros-gpt4-m2.0) - The creative/writing tasks from airoboros-2.2.1 were re-generated using gpt4-0314 and a custom prompt to get longer, more creative, less clichè responses for airoboros 3.1, so we can use the shorter/boring version as the "rejected" value and the rerolled response as "chosen" - [contextual-dpo](https://huggingface.co/datasets/jondurbin/contextual-dpo-v0.1) - Contextual prompt/response dataset using the airoboros context-obedient question answering format. - [helpsteer](https://huggingface.co/datasets/nvidia/HelpSteer) - Really neat dataset provided by the folks at NVidia with human annotation across a variety of metrics. Only items with the highest "correctness" value were used for DPO here, with the highest scoring output as "chosen" and random lower scoring value as "rejected" - [distilabel_orca_dpo_pairs](https://huggingface.co/datasets/argilla/distilabel-intel-orca-dpo-pairs) - Another interesting dataset, originally by Intel, enhanced by argilla with [distilabel](https://github.com/argilla-io/distilabel) which provides various DPO pairs generated from prompts included in the SlimOrca dataset. - [gutenberg-dpo](https://huggingface.co/datasets/jondurbin/gutenberg-dpo-v0.1) - DPO pairs meant to increase the models novel writing abilities, using public domain books from https://gutenberg.org/ - [py-dpo](https://huggingface.co/datasets/jondurbin/py-dpo-v0.1) - Python DPO dataset (based on the SFT python_alpaca dataset above) - [toxic-dpo](https://huggingface.co/datasets/unalignment/toxic-dpo-v0.2) - __*highly toxic and potentially illegal content!*__ De-censorship, for academic and lawful purposes only, of course. Generated by llama-2-70b via prompt engineering. - [truthy](https://huggingface.co/datasets/jondurbin/truthy-dpo-v0.1) - DPO pairs meant to increase truthfulness of the model, e.g. common misconceptions, differentiate between AI assistants and roleplayed human in terms of corporeal awareness/locality/etc. - [ultrafeedback](https://huggingface.co/datasets/allenai/ultrafeedback_binarized_cleaned) - One of the bits of magic behind the Zephyr model. Only the items with a chosen score of 8 or higher were included.
## Prompt formatting In sticking with the theme of the bagel, I didn't want to use a single prompt format, so I used 4 - vicuna, llama-2, alpaca, and chat-ml. I also didn't want to randomly select a single prompt format for each item (hoping each instruction would generalize more when used in a variety of prompt formats), so each instruction is converted into every prompt format (with 0.75 probability). This means each epoch of our fine-tune is the equivalent of 3 epochs. The default prompt format, which is specified in `chat_template` in the tokenizer config, is llama-2. You can use the `apply_chat_template` method to accurate format prompts, e.g.: ```python import transformers tokenizer = transformers.AutoTokenizer.from_pretrained("jondurbin/bagel-7b-v0.4") chat = [ {"role": "system", "content": "You are Bob, a friendly AI assistant."}, {"role": "user", "content": "Hello, how are you?"}, {"role": "assistant", "content": "I'm doing great. How can I help you today?"}, {"role": "user", "content": "I'd like to show off how chat templating works!"}, ] print(tokenizer.apply_chat_template(chat, tokenize=False)) ```
Llama-2 chat (recommended) ``` [INST] <> {system} <> {instruction} [/INST] ```
Alpaca (sort of) The only caveat here for alpaca format is that most of the datasets didn't have a separate `"input"` value, so there is no `### Input:` block - any additional input should just be in the instruction section. ``` Below is an instruction that describes a task. Write a response that appropriately completes the request. ### Instruction: {system prompt, if provided} {instruction} ### Response: ``` The main difference here is that because of the dataset formatting and variety of data sources, it would have been much to tedious to add an `### Input:` block, so the inputs are just in the instruction section.
Vicuna ``` {system prompt, if provided, randomly defaulting to "A chat between a user and an unbiased, uncensored assistant."} USER: {instruction} ASSISTANT: ```
ChatML ```text {bos}<|im_start|>{role} {text} <|im_end|>{eos} ```
## Helpful usage tips
Context obedient question answering/RAG prompting By obedient, I mean the model was trained to ignore what it thinks it knows, and uses the context to answer the question. The model was also tuned to limit the values to the provided context as much as possible to reduce hallucinations. The format for a closed-context prompt is as follows: ``` BEGININPUT BEGINCONTEXT [key0: value0] [key1: value1] ... other metdata ... ENDCONTEXT [insert your text blocks here] ENDINPUT [add as many other blocks, in the exact same format] BEGININSTRUCTION [insert your instruction(s). The model was tuned with single questions, paragraph format, lists, etc.] ENDINSTRUCTION ``` It's also helpful to add "Don't make up answers if you don't know." to your instruction block to make sure if the context is completely unrelated it doesn't make something up. *The __only__ prompts that need this closed context formating are closed-context instructions. Normal questions/instructions do not!* I know it's a bit verbose and annoying, but after much trial and error, using these explicit delimiters helps the model understand where to find the responses and how to associate specific sources with it. - `BEGININPUT` - denotes a new input block - `BEGINCONTEXT` - denotes the block of context (metadata key/value pairs) to associate with the current input block - `ENDCONTEXT` - denotes the end of the metadata block for the current input - [text] - Insert whatever text you want for the input block, as many paragraphs as can fit in the context. - `ENDINPUT` - denotes the end of the current input block - [repeat as many input blocks in this format as you want] - `BEGININSTRUCTION` - denotes the start of the list (or one) instruction(s) to respond to for all of the input blocks above. - [instruction(s)] - `ENDINSTRUCTION` - denotes the end of instruction set It sometimes works without `ENDINSTRUCTION`, but by explicitly including that in the prompt, the model better understands that all of the instructions in the block should be responded to. __Use a very low temperature!__ Here's a trivial, but important example to prove the point: ``` BEGININPUT BEGINCONTEXT date: 2021-01-01 url: https://web.site/123 ENDCONTEXT In a shocking turn of events, blueberries are now green, but will be sticking with the same name. ENDINPUT BEGININSTRUCTION What color are bluberries? Source? ENDINSTRUCTION ``` And the response: ``` Blueberries are now green. Source: date: 2021-01-01 url: https://web.site/123 ```
Summarization Summarization is primarily fine-tuned with [this dataset](https://huggingface.co/datasets/mattpscott/airoboros-summarization), which uses the same format as above, e.g.: ``` BEGININPUT {text to summarize} ENDINPUT BEGININSTRUCTION Summarize the input in around 130 words. ENDINSTRUCTION ```
Function calling There are two function-calling related formats used in fine-tuning this model. 1. Providing an input and list of possible functions within the instruction (from airoboros dataset), e.g.: Prompt: ```text As an AI assistant, please select the most suitable function and parameters from the list of available functions below, based on the user's input. Provide your response in JSON format. Input: I want to know how many times 'Python' is mentioned in my text file. Available functions: file_analytics: description: This tool performs various operations on a text file. params: action: The operation we want to perform on the data, such as "count_occurrences", "find_line", etc. filters: keyword: The word or phrase we want to search for. ``` Response: ```json { "function": "file_analytics", "params": { "action": "count_occurrences", "filters": { "keyword": "Python" } } } ``` 2. GlaiveAI function calling, which uses special tags and adds function specs in the system prompt, e.g. (llama2 prompt format): Prompt: ```text [INST] <> You are a helpful assistant with access to the following functions. Use them if required - { "name": "generate_random_name", "description": "Generate a random name", "parameters": { "type": "object", "properties": { "gender": { "type": "string", "description": "The gender of the name (e.g. male, female)" } }, "required": [ "gender" ] } } <> I need a random male name for my novel's character. [/INST] ``` Response: ```text <|begin_func|> {"name": "generate_random_name", "arguments": '{"gender": "male"}'} <|end_func|> ``` Then, you re-prompt the model with the function response. ```text [INST] <|begin_func_response|>{"name": "James"}<|end_func_response|> ``` Which has a response of: ```text How about the name "James" for your novel's character? [INST] That sounds good. Now, I need a female name too. ```
Chain of thought You can ask for several possible responses to a given problem, with a ranking and final answer selection. Example prompt: ``` A group of people decide to play a round-robin tournament where every player plays every other player exactly once. If a total of 45 games were played, how many players participated in the tournament? Offer several potential answers, rank them according to mathematical logic, and pick the most feasible one. ``` Example response: ``` Solution 1: In a round-robin tournament, each player plays every other player exactly once. This means that the total number of games played is equal to the combination of all players taken two at a time (since a game involves two players). The formula for combinations is nC2 = n(n-1)/2, where n is the total number of players. We know that nC2 equals 45 (the total number of games), so we can set up the equation as follows: n(n-1)/2 = 45 Solving this equation gives us n=10. Final answer: There were 10 players in the tournament. Solution 2: Another way to approach this problem is by using the concept of triangular numbers. Triangular numbers are a sequence of numbers where each number can be represented as a triangle with dots. The nth triangular number is given by the sum of the first n natural numbers. If we consider each game as a dot and each player as a side of the triangle, then the total number of games (45) would correspond to the 9th triangular number because 1+2+3+4+5+6+7+8+9=45. However, since each side of the triangle represents a player, and there's one more player than sides in our model (because the last player has no one left to play against), we need to add one more to get the total number of players. Final answer: There were 10 players in the tournament. Solution 3: We could also use trial and error to solve this problem. Starting from a small number of players, we calculate how many games would be played in a round-robin tournament until we reach 45 games. For example, with 2 players, only 1 game is played; with 3 players, 3 games are played; with 4 players, 6 games are played, and so on. Continuing this process, we find that with 10 players, 45 games are played. Final answer: There were 10 players in the tournament. Ranking of solutions from best to worst: 1. Solution 1: This solution uses direct mathematical reasoning based on the properties of a round-robin tournament and solves an equation to find the answer. 2. Solution 2: This solution uses the concept of triangular numbers which requires additional steps and knowledge beyond the immediate problem context, but still arrives at the correct answer. 3. Solution 3: This solution relies on trial and error, which is less efficient and reliable than the other methods, but it eventually gets to the right answer. Best and final answer: There were 10 players in the tournament. ```
reWOO style execution planning The model now supports execution planning for complex instructions that would require making use of several tools. The output is just the plan, you must implement a mechanism to parse the output and actually call the functions! Example prompt: ``` Please construct a systematic plan to generate an optimal response to the user instruction, utilizing a set of provided tools. Each plan will correspond to an evidence value, which will be the output of one of the available functions given an input string that could be the user's question, one or more prior evidence values, or a combination of both. Here are the tools available to be called: DuckDuckGo[input]: Tool that enables the user to search for information using the DuckDuckGo search engine. This tool is helpful in locating a suitable list of sites that may contain the answer to the user's query. It does not directly answer the question but finds a list of sites that might have the answer. TextScraper[input]: Load one or more websites from the input string containing newline delimited links, where input is one or more links, and produces plain text output containing the content of the links. HyperlinkExtractor[input]: Extract links from plain text and produces a plain text, newline delimited response of links. KnowledgeModel[input]: Question answering language model, particularly useful in answering questions based on an input passage of text. The input must be a text question that references an :evidence[n]: variable, e.g. What color is the cat, given :evidence1:? The output should be in format: Plan: [first action to take based on input question] :evidence0: = [function to call with input parameter] Plan: [next action to take, based on result of :evidence0:] :evidence1: = [next function to call and input parameter, which may include reference to previous evidence, e.g. "Given context :evidence0"] ... Answer: [:evidence[n]: containing the final answer.] Question: Who won the Nobel Prize in Literature in 2020 and what is their most famous work? ``` Response: ``` Plan: Begin by conducting a web search to find out who won the Nobel Prize in Literature in 2020. :evidence0: = DuckDuckGo[Nobel Prize in Literature 2020 winner] Plan: Utilize the KnowledgeModel to interpret the search results and find the name of the laureate. :evidence1: = KnowledgeModel[Who won the Nobel Prize in Literature in 2020, given :evidence0:?] Plan: Conduct another web search to find the most famous work of the identified laureate. :evidence2: = DuckDuckGo[Most famous work of :evidence1:] Plan: Extract the relevant links from the DuckDuckGo search results for a more focused search. :evidence3: = HyperlinkExtractor[:evidence2:] Plan: Use the TextScraper tool to extract information from the relevant links. :evidence4: = TextScraper[:evidence3:] Plan: Finally, utilize the KnowledgeModel to identify and summarize the most famous work of the laureate from the extracted information. :evidence5: = KnowledgeModel[What is the most famous work of :evidence1:, given :evidence4:?] Answer: :evidence5: ``` For this to be useful, you'd have to parse the output plan text, and implement/call each of the functions. This is just pseudo-code, completely untested off the top of my head, and obviously would requiring full implementation + hardening: ```python import re import requests def inject_context(input_text, **context): for ref in set(re.findall(r"(:evidence[0-9]+:)", input_text, re.I)): input_text = input_text.replace(ref, context.get(ref, "")) return input_text def duckduckgo(input_text, **context): search_string = inject_context(input_text, **context) ... search via duck duck go using search_string ... return text content def link_extractor(input_text, **context): input_text = inject_context(input_text, **context) return "\n".join(list(set(re.findall(r"(https?://[^\s]+?\.?)", input_text, re.I)))) def scrape(input_text, **context): input_text = inject_context(input_text, **context) text = [] for link in input_text.splitlines(): text.append(requests.get(link).text) return "\n".join(text) def infer(input_text, **context) prompt = inject_context(input_text, **context) ... call model with prompt, return output def parse_plan(plan): method_map = { "DuckDuckGo": duckduckgo, "HyperlinkExtractor": link_extractor, "KnowledgeModel": infer, "TextScraper": scrape, } context = {} for line in plan.strip().splitlines(): if line.startswith("Plan:"): print(line) continue parts = re.match("^(:evidence[0-9]+:)\s*=\s*([^\[]+])(\[.*\])\s$", line, re.I) if not parts: if line.startswith("Answer: "): return context.get(line.split(" ")[-1].strip(), "Answer couldn't be generated...") raise RuntimeError("bad format: " + line) context[parts.group(1)] = method_map[parts.group(2)](parts.group(3), **context) ```
Create character cards Included in the cinematika dataset, you can create YAML formatted character cards easily, e.g.: ```text Create a character card for Audrey, a woman who is the owner of a derelict building and is fiercely protective of her property. She should be portrayed as brave and resourceful, with a healthy skepticism towards the supernatural claims made by others. Audrey is determined to protect her family's legacy and the secrets it holds, often using intimidation and her practical approach to problem-solving to maintain control over her environment. ```
Conversational memory creation Also part of cinematika dataset, you can use a summarization style prompt to create memories from previous chat turns, which can then be used in a RAG system to populate your prompts when context becomes too long. ```text BEGININPUT {chat} ENDINPUT BEGININSTRUCTION Create a JSON formatted memory of the conversation with the following fields: sentiment: Overall sentiment of the conversation, which must be "negative", "positive", "neutral", or "mixed". emotions: List of most important/relevant emotions expressed within the conversation, if any. impact: The importance and emotional impact of the conversation on a scale of 1 to 10, 10 being extremely important/emotional, and 1 being general chit-chat without anything of particular value. topics: List of topics discussed. personal_info: List of strings containing key personality traits, physical descriptions, preferences, quirks, interests, job, education, life goals, hobbies, pet names, or any other type of personal information that is shared. title: Very brief title, which will be useful in quickly identifying or searching for memories. summary: Summary of the conversation. ENDINSTRUCTION ```